After the first conversation

A way of thinking, shared.

Rod, Bryce, Marvin: thank you for the hour on the 6th. Before the working session, we wanted to set down what we heard, plainly, so you can correct it. This is still not a proposal and not a price. It is our read of the real question, which sounds less like “can AI replace an assistant” and more like “where can the firm gain dependable leverage without spending down the trust it sells.” Read it, argue with it, keep what is useful. There is no obligation on the other side of it.

How this works

On Listen, Frame, and Envision you are invited to react, re-rank, and edit what you see, so the working session starts from your corrections rather than our guesses. The Article lays out our point of view in full if you want the longer read, and Candidate workflows is the shortlist we would narrow to one together. Whatever you affirm gathers, quietly, into one page you can bring with you.

Thursday’s working session

One session, and one thing chosen

{{ meetingDate }}  ·  Rod, Bryce, Marvin, Nick, and Eric

This one is hands-on. We play back what we heard and you correct it, we look together at the candidate workflows and pick one worth a small first proof, and I show a private, local model running on hardware you would own, so nothing privileged leaves the building. We leave with one bounded workflow chosen, its standard of done, and who owns each step. Nothing here is fixed. If a thread matters more than the clock, we follow the thread.

We are not here to decide whether AI replaces an assistant. We are here to make the hidden work visible, then decide which parts stay human, which can be systematized, and which a bounded agent can safely hold, and to prove it on one thing first.

Listen→Frame→Envision→One workflow→A page to keep
A Dark Horse Works point of view

The Chief of Staff You Couldn’t Afford, and the One You Can Now Build

Why the most valuable role in a modern firm was, until recently, reserved for the few, and what changes when the administrative floor beneath it can finally be held by something other than a person.

By Dark Horse Works/For Fuller IP Law/8 min read

#{{ sec.heading }}

{{ p.text }}

{{ p.text }}

{{ p.text }}
A Dark Horse Works point of view
Listen

What we heard

Sorted by whose it is, and marked by where it comes from: said to us on the call, structural, or still unspoken and inferred. Most of this we now heard first-hand. Affirm what lands, correct what doesn’t, set aside what we should. Open the fuller context whenever it helps.

{{ listenProgress }}
reviewed
Where we start

Here is what we think we heard on the 6th. We had the hour with all of you, so most of this is first-hand now, not a guess, but our read can still be wrong in the details, and the most useful thing you can do before the working session is tell us where. The question we heard sounds less like “can AI replace an assistant” and more like: where does Fuller IP want dependable leverage, and where would automation create unacceptable risk because the work depends on confidentiality, judgment, or relationship context? What did we get right, what did we miss, and what would you say differently?

Whose
{{ listenStepLabel }}
{{ listenCurrent.num }} {{ listenCurrent.whoLabel }} {{ listenCurrent.layer }} · {{ listenCurrent.layerNote }}

{{ listenCurrent.headline }}

{{ listenCurrent.body }}

{{ listenCurrent.detail }}
What did we miss
Frame

Trusted leverage, and the turn

First the crux, then a few ways to hold the problem: re-rank them by which feels truest, and say whose problem each is. Then the move the rest of this rests on.

The crux

One bounded proof, not a broad program.

The conversation on the 6th settled the question of whether AI can help here. The harder discipline now is resisting the urge to do everything at once. Much of what makes support valuable is tacit: knowing what matters, what can wait, what needs a human touch, when to escalate. The work is to make that hidden layer visible, then prove the idea on one bounded workflow against a standard you already hold, and let scope and trust grow from a real result rather than a plan.

3A

The problem, a few ways to hold it

Reorder by which feels most true to the firm, by dragging a row or using the arrows. Assign whose problem each is.

{{ p.rank }}
drag
{{ p.code }} · {{ p.source }}

{{ p.text }}

How we read it: {{ p.read }}
Whose problem is this
A way to sort the work

Not human or machine. Three places the work can live, once it is visible.

{{ b.label }}

{{ b.items }}
3B

The turn

What most firms hear

AI replaces the assistant. The firm re-hires, or it doesn’t, and either way the work shrinks to a line on a spreadsheet.

The truer frame, and you said it first

The agent promotes the human.

You framed it as AI the power tool, amplifying the craftsman. Same move, one level up: when a bounded agent holds the administrative floor, everyone climbs. The partner rises to strategy and the client craft only he can do; the person who ran the floor rises to orchestration and judgment; the associate rises to growth work. The roles do not shrink. They grow.

Why this isn’t “AI replaces a person”

The split between chief of staff up here and assistant down there was never a law of nature. It was a limit of human bandwidth: no one person could hold the relentless floor and rise to the strategic work at the same time. Redistribute the floor to a bounded system and the limit lifts. This is not removing a person. It is preserving the right human attention for the moments where trust and judgment actually matter.

The frame we’d start from

Preloaded, but yours to edit. It commits to nothing. It is just the sentence the rest would start from.
Envision

Choose the destination

Before we solve anything, pick which future is worth the change. These are not the same destination. Choose one, then choose the elements that describe where this should go.

{{ destinationNote }}

What it looks like, ninety days in

Choose the elements that describe where this should go. Edit any that are close but not right.

{{ e.tagLabel }}

{{ e.title }}

{{ e.body }}
The trade-offs to choose
{{ t.pair }}
{{ t.q }}

If this worked beautifully but did not immediately reduce headcount, would it still be worth doing?

What we are not claiming

A clear argument earns trust by naming its own limits. Edit these, or add your own.

{{ na.text }}
Passages you marked on the article

{{ fp.text }}
The choice

When the floor can be automated, a firm chooses: eliminate the role, or elevate the person.

The tools are identical. The firms are not. The question isn’t which vendor, or even how much it costs. It is what kind of firm you intend to be, and whether the leverage you gain protects the trust you sell.
Explore

A map to argue with

Before we decide what to build first, we mapped the candidate workflows against the dimensions that actually determine a good first proof: how much senior time it reclaims, how safely it can be trusted near client work, how measurable it is in ninety days, and how cleanly it runs behind your own firewall. Two candidate sets share the same map: your IP law-firm workflows (circles) and the EOS operating-cadence workflows (diamonds). Toggle either set, or any blast-radius band, on and off. This is a map to argue with, not a recommendation to accept.

Dataset
Axes
Blast radius How much damage if the agent gets it wrong. Green is safe to try; red is high-exposure.
{{ plotSvg }}
How to read this

The best first proof sits up and to the right, runs often (a larger marker), and is green: high value, safe to fail, frequent, private. The flashiest work, patent drafting and live meeting capture, sits lower or redder. Real upside, wrong first step. Positions are relative rank on each axis, so no two workflows sit on top of each other; the detail card shows the underlying one-to-five scores.

{{ activeBubble.zoneLabel }} · {{ activeBubble.dsLabel }}

{{ activeBubble.name }}

What it is

{{ activeBubble.what }}

Why it matters

{{ activeBubble.why }}

For example

{{ activeBubble.example }}

{{ sc.label }} {{ sc.val }}
The metric it would move {{ activeBubble.metric }}
Read a workflow

Hover or tap any workflow on the map to see what it is, why it matters, an example, its seven scores, and the one metric it would move.

Circles are IP law-firm workflows; diamonds are EOS operating-cadence workflows.
Where we’d suggest starting

Where the map points for Fuller is a matter-flow and open-loop copilot: it pulls matter, deadline and hand-off status from the firm’s own systems into a weekly at-a-glance, with red-flags and follow-up accountability. It answers the “who’s got the potato” hand-off pain and the dropped-follow-up pain directly.

It lives in the green zone of both maps. It needs no recording of client conversations. And it can show a real before-and-after number in ninety days. Offered as where we’d suggest starting, to argue with, not to accept.
The first step

Candidate first workflows

A shortlist of bounded workflows, each small enough to prove without heroics and relevant enough to matter. Three we sketched with you live on the 6th; two more surfaced from the map on Explore. We would narrow this to one at the working session. Mark the ones worth proving first; what you choose gathers on the page you keep.

The discipline

One bounded workflow. A clear standard of done. A named owner for each step. A first win earns the next; nothing here has to be a leap of faith.

{{ w.key }} {{ w.sourceLabel }}
Worth proving ✓

{{ w.title }}

{{ w.produces }}

Standard of done{{ w.standard }} Why it’s a good first step{{ w.firstBecause }}
{{ sp.label }}

This list is deliberately short, and it will grow as we learn more, toward a handful of real candidates. The point of the session is not to admire the menu. It is to choose one.
If it’s useful

How we’d work together

We can give you a sense of how engagements like this usually work, but the immediate next steps are simpler and concrete. We’d rather earn scope with a real result than scope it on a slide. So the plan is a mutual NDA, then a working session that’s on us, then a single workflow proven before anyone talks about a larger commitment.

The immediate next steps
  1. {{ ns.num }} {{ ns.title }}{{ ns.note }} {{ ns.who }}
How engagements like this unfold
{{ o.key }} Start here

{{ o.title }}

{{ o.body }}

Timeline {{ o.timeline }} Best when {{ o.bestWhen }}

The working session above is the honest first step, and it’s free. These are simply how things tend to unfold after it, once we all know what we’re building.

Engagement flow

A staged progression from diagnosis to controlled proof to capability build. Each gate is binary, and its off-ramp is a real answer, so we never drift into permanent-pilot syndrome.

The decision gates
  1. {{ g.label }} {{ g.q }}

The gates are managerial, not technical: baseline set, scope approved, controls documented, KPI delta observed, owner assigned.

Before a real proposal

We’d want to answer five questions first, so we neither over- nor under-scope. Scope, and therefore price, follows the shape of the work, so we’d rather discuss it once the shape is clear.

  1. {{ q.num }} {{ q.text }}
A page to keep

What we explored together

This gathers only what you affirmed or wrote across the pages, so you can bring it into the working session. It is a keepsake of the thinking, not a proposal and not an agreement. Take it with you, or forward it on.

Dark Horse Works · for Fuller IP Law

What we explored together

{{ sessionDate }}

As you move through the pages, what you affirm will gather here.
The frame we’d start from

{{ memoFrame }}
The destination we’re aiming at

{{ memoDestination.key }} · {{ memoDestination.label }}. {{ memoDestination.vision }}
What we heard that landed
  • {{ m.headline }}. {{ m.body }} · {{ m.who }}
    •
How we’re holding the problem
  1. {{ m.text }} · {{ m.owners }}
    2.
The destination we pictured
  • {{ m.title }}. {{ m.body }}
    •
Workflows worth proving first
  • {{ m.title }}. {{ m.produces }}
    •
What we are not claiming
  • {{ n.text }}
    •
Passages you marked

{{ fp.text }}
Dark Horse Works
If the thinking is useful, let’s talk.
No commitment. A conversation, whenever it’s useful to have one. {{ contactLink }}
Start a conversation